Agentic search: what it means for your webshop traffic

Agentic search: fewer visitors, more purchase intent

Agentic search means an AI takes over the search work from your customer. Your customer no longer scrolls through ten webshops; an agent searches, compares and returns a shortlist. That sounds like the future, but it is already visible in the measurement data. Adobe analysed more than a trillion visits to US retail sites and saw traffic from AI assistants grow 138% year over year in May 2026. The same analysis shows those visitors convert 54% better than visitors from other channels.

For you as a webshop owner, this means two things. One: a channel is growing that converts better than anything you currently have. Two: that channel looks at your webshop in a completely different way than a human does. An agent sees no beautiful design and feels no atmosphere. It reads data, compares claims and makes a recommendation. If your product information is unreadable to it, you simply do not exist in that channel.

What exactly is agentic search?

Agentic search is search where an AI agent performs tasks on behalf of the user: searching, filtering, comparing and, in its final stage, even paying. The difference with a classic search engine is who does the work. With classic search, your customer types a query and clicks through the results themselves. With agentic search, the customer gives an assignment, for example "find a desk chair under two hundred euros that fits a small office", and the agent does the rest.

In practice, agentic search consists of three layers, each taking over a different part of the customer journey:

The third layer is furthest away from most Dutch webshops, but the first two already run at volume today. How AI agents complete purchases via Shopify and which protocols sit underneath is covered in our article on agentic commerce via Shopify. This article is about the layer before that: what happens to your traffic once agents take over the search work.

The word "traffic" takes on a different meaning here. For years, traffic was the goal: more sessions, more visitors, more clicks. In an agentic channel, traffic is a consequence. The agent does the research, and only the customer with serious purchase intent still clicks through. That changes what you should optimise, and above all: where.

What shifts compared to classic SEO?

The reflex of many webshop owners is to treat agentic search as yet another Google update: grab the checklist, make a few tweaks, done. That reflex does not apply here. Agentic search does not change the rules within the game, it changes who is playing. Your page is no longer judged only by a human who scans and clicks, but by a machine that reads, compares and summarises.

Two things stand out in that table. First: the assets overlap. A page that is complete, factual and structured scores in both worlds. There is no choice between "doing SEO" and "getting agent-ready", there is only a stricter test on work you were supposed to be doing anyway. Second: the measurement point shifts. Positions and sessions say less and less about what you sell. A product that appears in no AI recommendation at all can sit at position three in Google and still structurally miss revenue from the audience that no longer arrives via Google.

That makes agentic search not a replacement for your search strategy, but a second layer on top of it. The foundation stays. The test gets tougher.

Fewer visitors who buy more: the new traffic mix

The first fear of every webshop owner on this topic: am I losing my traffic? The honest answer is that the top of your funnel gets thinner. Questions like "which desk chair works for a small office" used to produce a visit to your category page. An AI now answers that question itself, with your content as a source or without you. The click that remains falls later in the customer journey.

But the visitors who do click through are different visitors. Adobe saw that visitors from AI assistants stay 53% longer on the site and view 23% more pages than average. The explanation is simple: the agent already did the groundwork. Whoever arrives via an AI recommendation has already compared their options and comes to check whether the agent's promise holds up.

The traffic mix shifts from many loose research visits to fewer, but heavier visits with purchase intent. Whoever only looks at sessions sees a dip and draws the wrong conclusion. Whoever looks at revenue per visitor sees that the channel that is growing is also the channel that performs best. Compare it to a physical shopping street: fewer people walk in just to browse, but whoever walks in has already decided at home what they came to get. No shopkeeper complains about that. You do make sure your shop window matches what is inside.

For your measurement, this means you want to see AI traffic separately. In your analytics you recognise it by referring domains such as chatgpt.com and perplexity.ai. Put that group next to your organic traffic and compare it on conversion and revenue per visitor, not on volume. Within a month you will see whether the pattern from the Adobe data holds for your shop, and what each AI referral is actually worth. How to set that up is covered in our article on preparing your webshop for AI search results.

What an agent cannot read on your product page

The same Adobe analysis measured per category how much of the product content on retail sites is readable for AI systems. For cosmetics that is 63%, for furniture only 47%. That number is precisely the pain point: one of the best-converting channels of this moment cannot read roughly half of the average product page. There is no reason to assume Dutch webshops do better; they are built with the same platforms and the same export settings.

What makes a product page unreadable for an agent? In practice we see four recurring causes:

1. Specifications inside an image. That beautiful infographic with sizes and materials is a blank surface to an agent, unless the information also sits on the page as text.
2. Content that only loads after JavaScript. Many AI crawlers read the bare HTML. If your product description lives in a tab that loads client-side, that text does not exist for the agent.
3. Missing or empty alt texts. For you, the image carries half the story, but without a description the agent has no idea what it shows.
4. Product data without structure. No structured data, or structured data that deviates from what is visible on the page. An agent compares both and drops you when in doubt.

A human sees a polished product page. An agent sees a half-empty sheet. Closing that gap is not a design project but a data project, and it starts with your category and product pages. The seven elements that make the difference there are laid out in GEO-ready category pages.

Your feed is your new shop window

There is one place where the shift becomes most concrete: your product feed. ChatGPT Shopping does not show products based on how nice your website looks, but based on the feed you supply. OpenAI's Product Feed Spec says it literally: products come in via a media list, and the first entry in it is treated as primary. The first image in your feed is your shop window in ChatGPT.

The image order in your feed, something many webshops never consciously set, now decides which image represents your product in an AI recommendation. If a mediocre supplier photo or a random colour variant sits there, that is your first impression in a channel that converts better than your average visitor. The metadata around it counts too: alt text, width and height belong with every image in the feed.

This is exactly why we treat image order as a design decision, not an export setting. Strongest image first. A packshot that makes the product instantly recognisable, followed by images that show the situation it is used in. Deciding which product images can carry that weight is the craft of product photography, and the other signals ChatGPT weighs before recommending are covered in how ChatGPT recommends your product.

Whoever gets their feed in order wins twice. The feed powers Google Shopping, and that same feed is the source for AI assistants. One improvement, two channels.

What does Google itself say about visibility in AI search results?

Google published its own document on how sites end up in AI Overviews and AI Mode, and the core is refreshingly sober: there is no special markup for AI features. No secret tag. No llms.txt file that puts you at the front of the queue. Anyone selling you an "AI optimisation trick" built on that is selling air.

What Google does explicitly mention is remarkably concrete for webshops. Support your text with good images and videos. Use structured data that matches exactly what is visible on the page. And keep your Merchant Center feed up to date, because product information in AI answers comes from there. Three points, and all three are about the same principle: the machine must be able to confirm what the human sees.

That is good news for everyone who already took the fundamentals seriously. The webshop that describes its products completely, keeps its data in order and uses images that carry the story does not need to learn a new trick for agentic search. The webshop that leaned on loose tactics for years has a problem that grows with every AI update. Our approach to those fundamentals is in GEO for webshops, and for the technical side we look along through our SEO service.

bol is building a buying agent too

Anyone who thinks this is an American story that will only reach the Netherlands in a few years should look at bol. The platform is working on its own Shopper Agent and is already asking partners for richer content: images that show the product in different situations and data that is easy for AI to read. In that same publication, bol notes that 57% of Dutch consumers use AI tools when researching products.

The biggest sales channel in the Netherlands is preparing its own agent layer and telling sellers which content that layer needs. The sellers who enrich their listings now will be at the front when that agent goes live. The sellers who wait until it is unavoidable start with a handicap in a system that runs on historical data.

An interesting detail: bol's quality score only measures service standards, such as delivery promise and returns handling. Your images and product content do not count in it. The image track runs entirely via the AI side: the Content Uploader that structures your data and the Shopper Agent that is on its way. Whoever only steers on their score is optimising for half of the system and missing the half where the recommendations will come from.

For bol sellers the translation is practical. Situational images next to the standard packshot on white, so the agent can match on usage context. Complete, structured product data via the Content Uploader. And a listing built around claims an AI can understand and pass on, not around a pile of keywords. How to build such a listing is in optimising your bol.com listing, and why an AI reads your bullets as claims instead of keywords is in optimising your Amazon listing for AI. The mechanics are the same on every platform.

The agent selects on data, the human buys on trust

With all this technology it is easy to forget who decides in the end: a human. The agent makes the shortlist, but the human picks from it. And that human is younger and trusts AI more than you might think. Research by Bazaarvoice among more than eight thousand consumers shows that 41% of shoppers between 18 and 34 search for products via generative AI instead of a search engine, and that 75% of that age group trusts the recommendations that come out of it.

At the same time, that same consumer is sharp on brands that cut corners with AI themselves. Research by Meltwater and YouGov among ten thousand consumers shows that 86% think AI content should carry a label, and that 32% trust a brand less after AI content, versus 15% who trust it more. AI content costs a brand net trust the moment the customer notices.

The combination of those two movements is the core of this whole story. The agent selects you on data. The human buys on trust. You need both: product information a machine can read plus proof that convinces a human. Real reviews. Real product photography that shows what the customer gets. A brand that tells the same story on every channel. Whoever neglects one of the two loses: unreadable data means you never make the shortlist, and a bare brand means you get skipped on that list.

There is a third player in this game: your reviews. An agent that compares products reads not only your claims but also what buyers say back. A listing that promises what the reviews contradict falls through with a machine that puts both sources side by side. That may be the healthiest change of all: agentic search rewards products that deliver what they claim, and punishes the gap between marketing and reality without mercy.

What we do differently at Oase now that AI reads along

For six years we have been building listings, product pages and brands for e-commerce at Oase Creative. The work itself has not changed because of agentic search, but the bar has. Three things we now do structurally differently than two years ago.

First, we write product content as claims, not keyword lists. An agent reads "fits desks up to 80 centimetres and monitors up to 10 kilos" and can match that against the customer's question. With a row of loose keywords it can do nothing. That sounds small, but it flips the entire structure of a listing: first decide which five claims your product must make, then write.

Second, we treat images as data. Every product photo gets a job (recognise, understand, trust), a descriptive alt text and a deliberate place in the order, from the feed to the listing. With Screenmate we ran the complete brand journey, from nameless test product to a brand with one consistent visual line on every channel. That consistency is exactly what an AI system responds to: the same name, the same claims and the same images on the webshop, the marketplace and in the feed make your brand recognisable to a machine.

Third, we build the content layer around it. An agent that makes a recommendation looks for confirmation in sources. A knowledge base that seriously answers your customer's questions makes your product quotable at the moment the agent assembles its shortlist. Why that content layer is the best salesperson a webshop can hire is in blog content for your webshop. For the listing work itself, there is our product listing design service.

How to make your webshop agent-ready this month

Not a big project, but an order of operations. The temptation with shifts like this is either wanting everything at once ("we must become AI-proof") or doing nothing ("we will see when it gets here"). Both wrong. The numbers above are about traffic that is already arriving and converting today, so waiting costs revenue. But most of the work is simply overdue maintenance on your product data, not a new specialism. This is what you can do in four weeks, in the order in which it pays off:

1. Measure your AI traffic. Create a segment in your analytics for referrals from chatgpt.com, perplexity.ai and similar domains. You cannot steer what you cannot see.
2. Check what an agent reads on your product page. Turn JavaScript off and look at your most important product page. Everything that disappears does not exist for a large share of AI crawlers.
3. Get specifications out of your images. Every size, material and property that now only lives in an infographic also goes on the page as text.
4. Put your strongest image first in your feed. Open your product feed and check per top product which image sits at position one. That image is your shop window in ChatGPT Shopping.
5. Fill your alt texts. Descriptive and factual, no keyword stuffing. The agent uses them to understand what your image shows.
6. Align your structured data with your visible content. Price, stock and specifications in your data must match exactly what is on the page.
7. Rewrite your top five products as claims. Five concrete, checkable statements per product, each tied to a usage situation.

Whoever walks through these seven steps is ahead of the vast majority of Dutch webshops. Not because the steps are hard, but because almost nobody takes them as long as traffic still comes in "as usual" through the old channels. The numbers above show that head start is temporary.

Not sure where your webshop stands? We audit product pages, feeds and listings and tell you exactly what an agent does and does not see of your product. Book a call or first have a look at our SEO approach.

Agentic search takes over the search work, not the purchase decision. The webshops that win are the ones that take both seriously: readable for the machine, convincing for the human. That is not a new discipline. It is the old craft, finally enforced.